DeepSeek releases DeepSeek-V4-Flash model

DeepSeek-V4-Flash 0731 delivers 532 solves per $100 on DeepSWE, costing roughly one-sixth of GPT-5.6 Luna. A cascade strategy using the model first and escalating to Luna on failure solves 78.9% of tasks at a 37% lower cost than using Luna alone.
How this story unfolded
4 weeks · 16 reports · 91 community posts · 107 of 113 shown
- Jul 16
- Jul 17
- Jul 20
- Jul 31
DeepSeek Unveils Public Beta API for Flagship AI Modelbloomberg.com
DeepSeek puts V4-Flash API into public betatechnode.com
deepseek-ai/DeepSeek-V4-Flash-0731huggingface.co
unsloth/DeepSeek-V4-Flash-0731-GGUFhuggingface.co
DeepSeek V4 Flash now runs updated weights on AI Gatewayvercel.com
DeepSeek Upgrades DeepSeek-V4-Flash-0731 with Major Agentic and Coding Gainsmarktechpost.com
- Aug 1
- Aug 2
- Aug 3
- Aug 4
DeepSeek-V4-Flash API launches on China’s National Supercomputing Internettechnode.com
huihui-ai/Huihui-DeepSeek-V4-Flash-0731-abliterated-GGUFhuggingface.co
DeepSeek V4 Flash: The Cheapest Frontier-Level Open Model Yetmindstudio.ai
How to Run DeepSeek V4 Flash Locally: Hardware, Quantization, Testsmindstudio.ai
DeepSeek V4 Flash is 90% off through Novita on AI Gatewayvercel.com
- Aug 5
- Aug 6
- Aug 7
- Aug 8
- Aug 9
- Aug 10
- Aug 11
- Aug 12
DeepSeek by email
Get an email when DeepSeek has news
No news that day, no email.
More stories today
- MiniMax H3 model integration on Fal to be showcased in live session
- Alibaba releases open weights for 2.4T-parameter Qwen3.8-Max model
- NVIDIA Spectrum-X Ethernet Photonics now in full production
- Agent evals are stuck in the chatbot era, says Raindrop's Ben Hylak
- GitHub blog details strategies for managing AI-generated pull requests